Download presentation
Presentation is loading. Please wait.
Published byCarson Horace Modified over 9 years ago
1
D EPT. OF I NFO. & C OMM., GIST Adaptive Dual-layered Bucket Synchronization Control for Distributed Virtual Environment JaeYoun Kim, Seokhee Lee and JongWon Kim 2005/8/23 {jykim, shlee, jongwon}@netmedia.gist.ac.kr Networked Media Laboratory Dept. of Information & Communications Gwang-Ju Institute of Science & Technology APAN NRW 2005
2
D EPT. OF I NFO. & C OMM., GIST Contents Background Related Work Motivations & Contributions Overall Framework Simulation & Experiment Results Conclusion
3
D EPT. OF I NFO. & C OMM., GIST Background In these days Advances in computer graphics, parallel/distributed system and high-speed networking Demands on building distributed virtual environments (DVEs) system Consistency Ensure consistent view for each participant over time-varying Network Tradeoff between responsiveness and consistency Synchronization algorithm Playout delay control
4
D EPT. OF I NFO. & C OMM., GIST Playout Delay Control Adaptation efficient playout delay Event Generation Event Execution Small Playout Delay Large Playout Delay Large Network Delay Network delay Synchronization delay
5
D EPT. OF I NFO. & C OMM., GIST Related Work Y. Ishibashi et al. Enhance a virtual-time rendering (VTR) algorithm for DVEs. The synchronization maestro When the network load is heavy, gradually increases (i.e., delays) the target output time. When the network load becomes lighter, gradually decreases the target output time. D. Lee et al. Add dynamic playout delay control scheme in bucket synchronization. Current network state is estimated by measuring how many events have been lost.
6
D EPT. OF I NFO. & C OMM., GIST Motivations and Contributions Motivations Event loss rate can not guarantee correct estimation of network state. Event loss when decreasing the playout delay because of skipping. Contributions Adaptive dual-layered bucket Estimates network state based on one- way transmission delay and event loss rate. Decrease the playout delay for silence period in order to prevent from skipping (event loss).
7
D EPT. OF I NFO. & C OMM., GIST Synchronization System Model System model All participants use a multicast address for event and report (T max ) Each participant is based on bucket-based synchronization layer Adaptation master among participants determine reference playout delay (R delay )
8
D EPT. OF I NFO. & C OMM., GIST Reference Playout Delay Adaptation master Group maximum transmission delay (G max ) Maximum value of T max Smoothed current G max ( C delay ) Filter out short-term fluctuations caused by jitter Upper target threshold (UTT), Lower target threshold (LTT) Reference playout delay based on UTT and event loss rate
9
D EPT. OF I NFO. & C OMM., GIST Dual-layered Bucket Management Each participant Receives the new playout delay (R delay ) from adaptation master. However, does not apply it directly. Prevent event loss when decreasing the playout delay If there is sending idle time like silence period in DVE systems, apply the playout delay. Virtual layer: provide guideline for possible playout delay Candidate playout delay (CR delay ) Real layer: detect sending idle time like silence period Established playout delay (ER dealy )
10
D EPT. OF I NFO. & C OMM., GIST Dual-layered Bucket Management(cont’)
11
D EPT. OF I NFO. & C OMM., GIST Simulation and Experiment Simulation Generate CBR traffic and TCP traffic (FTP) in NS-2 DVE systems( 4 nodes) communicate through single multicast address Compare playout delay Experiment Generate background UDP traffic (20Mbps~70Mbps) by iperf DVE systems( 2 nodes) communicate through multicast router(CISCO 2600) Compare skipping rate Topologies
12
D EPT. OF I NFO. & C OMM., GIST Simulation Result Network state estimation based on delay and loss (Proposed scheme) Network state estimation based on loss rate Proposed scheme Provides more correct estimation of current network state Guarantees responsiveness and interactivity of DVEs
13
D EPT. OF I NFO. & C OMM., GIST Experiment Result Comparison of playout delay Similar to simulation result Event loss rate Proposed scheme provides smaller loss rate
14
D EPT. OF I NFO. & C OMM., GIST Conclusions Dual-layered Bucket Management Provides more correct network state estimation to determine playout delay. Reduces event loss in event buffer for consistent global state. Future work Improve the proposed scheme for haptic-based DVEs. Haptic interface : 1 kHz I/O rate. Haptic interaction : more sensitive to delay and jitter (about 30 ~ 60 ms). Current scheme can not guarantee consistency in haptic-based DVEs.
15
D EPT. OF I NFO. & C OMM., GIST Thank You… Question ?
Similar presentations
© 2024 SlidePlayer.com Inc.
All rights reserved.